Ann Majewicz, PhD1, Marzieh Ershad1, Robert Rege, MD2. 1University of Texas at Dallas, 2UT Southwestern Medical Center
Great musicians, all-star athletes, and highly skilled surgeons share one thing in common: to the casual observer, the appearance of the expert in action can be described as fluid, effortless, swift, and decisive. Given that our understanding of expertise is so innate and engrained in our vocabulary, we seek to evaluate the effectiveness of using simple semantic descriptors to coach surgical residents as they acquire new surgical skill.
We are developing an autonomous coaching system that will present descriptive words to a surgical trainee as they practice tasks on a surgical simulator. The key features of this system include a lexicon of contrasting adjectives that can differentiate expertise (e.g., fluid-viscous, calm-anxious, slick-clumsy, etc.) and methods for correlating the sematic descriptors to movement signatures or user physiological response. We hypothesize that trainees who are exposed to the more natural and intuitive coaching vocabulary will achieve proficiency, and perhaps even expertise, more quickly than trainees who do not receive data-driven verbal feedback.
In preliminary work, we began developing a lexicon of contrasting adjectives to describe surgical expertise. We chose the word pairs: “viscous-fluid”, “rough-smooth”, “crisp-jittery”, “sluggish-swift”, “anxious-calm”, and “tense-relaxed”. One novice and one intermediate surgeon preformed two tasks on the da Vinci Standalone Simulator. Video snapshots of the user posture and the simulated task were posted on Amazon Mechanical Turk. 80 crowdworkers were asked to select words for each video. The novice user was assigned positive words approximately 35.7% of the time, and the intermediate was assigned positive words 61.7% of the time. Overall, both the intermediate and the novice received higher ratings for the “swift-sluggish” word pair and lower ratings for the “smooth-rough” word pair. This suggests that movement-based words may be a better metric for expertise than time-based words. These preliminary results were presented as a late-breaking poster at the 2015 IEEE ICRA conference.
Currently, we evaluating algoirthms which correlate movement signatures and user physiological response to the semantic descriptor pairs. We are using a variety of sensors to collect these data, including (1) inertial measurement units to record linear and angular accelerations of the limbs, (2) electromagnetic trackers to measure joint movement, and (3) physiological sensors to record muscle activity, heart rate, and skin galvanic response (measures of physical effort and stress). Our data-collection system simultaneously captures sensor data as well as video for simulated surgical tasks. We have collected data for 5 surgical residents (PGY1-4), performing a manipulation task and a needle driving task on the stand alone simulator. Data analysis is currently undeway and we expect to recruit a total of 12 residents (3 per residency year), as well as three expert surgeons to participate in this study. The data analysis results will indicate whether or not movement signatures can be correlated to words of expertise, thus supporting our hypothesis of building a data-driven, word-based autonomous coaching system for meaningful surgical skill training.